使用有效听觉编码的噪声抑制算法  

A noise reduction approach with efficient auditory coding

作　　者：张戈亮[1] 朱莉[1] 胡广书[1] 

机构地区：[1]清华大学生物医学工程系,北京100084

出　　处：《北京生物医学工程》2013年第2期111-116,共6页Beijing Biomedical Engineering

基　　金：国家自然科学基金(30970756);清华裕元医学科学基金(20200521)资助

摘　　要：目的噪声抑制是语音信号处理的重要课题。本文利用有效听觉编码算法处理语音信号,研究其在噪声抑制问题上的效果。方法基于有效听觉编码算法对于语音信号和白噪声信号在编码系数大小分布上具有显著差别的特性,本文提出一种使用有效听觉编码的降噪算法,在有效听觉编码算法获得的编码系数中抑制噪声,适用于含有加性白噪声的语音信号。结果客观评价结果表明本算法能够有效降低语音中的加性白噪声。同时将该算法与其他当前基于短时傅里叶变换的语音降噪算法进行了对比,表明该降噪算法对于信噪比(signal-to-noise ratio,SNR)的提高比当前得到广泛采用的最小均方误差算法(minimum mean-square error,MMSE)、谱减法等性能更好。结论本文提出的噪声抑制算法可以进一步提高当前噪声抑制算法的性能。Objective Noise reduction is an important topic in speech signal processing. This paper, which is motivated by the efficient auditory coding method, applies this method on suppressing noise in speech. Methods This paper finds that significant differences exist in the distributions of coding coefficients between speech signal and white noise with efficient auditory coding. This fact could be utilized in separating speech from noise and a noise reduction approach is proposed here to be applied in speech with additive white noise. Results Objective criterion shows that our method could remove additive white noise effectively. We also compare this method with other two noise reduction methods based on short-time Fourier transform. The results indicate that the proposed method could result in better signal-to-noise improvements （SNRI）than minimum mean square error（MMSE） method and spectral subtraction method which are widely used. Conclusions The proposed method makes improvement on current noise reduction methods.

关 键 词：有效听觉编码 噪声抑制 最小均方误差 语音信号 白噪声 

分 类 号：R318.04[医药卫生—生物医学工程] TN911.72[医药卫生—基础医学]